Re: [RFC PATCH 07/38] arm_mpam: resctrl: Add boilerplate cpuhp and domain allocation

[Jonathan Cameron <jonathan.cameron@huawei.com>]

> >> +static struct mpam_resctrl_dom *
> >> +mpam_resctrl_get_domain_from_cpu(int cpu, struct mpam_resctrl_res *res)
> >> +{
> >> +	struct mpam_resctrl_dom *dom;
> >> +	struct rdt_ctrl_domain *ctrl_d;
> >> +
> >> +	lockdep_assert_cpus_held();
> >> +
> >> +	list_for_each_entry_rcu(ctrl_d, &res->resctrl_res.ctrl_domains,
> >> +				hdr.list) {
> >> +		dom = container_of(ctrl_d, struct mpam_resctrl_dom,
> >> +				   resctrl_ctrl_dom);  
> > 
> > I'm lazy so haven't checked for more code here in later patches, but
> > if not, why not iterate the list to access the domain directly rather
> > than jumping through the rdt_ctrl_domain?
> > 
> > Something along lines of:
> > 	
> > 	list_for_each_entry_rcu(dom, &res->resctrl_res.ctrl_domains,
> > 				resctrl_ctrl_dom.hdr.list) {
> > 	}
> >  
> 
> Unless I've misunderstood I don't think this works because it's not what
> the fs/resctrl code expects.

I think I explained this one badly.

This should be functionally identical to the above so no visible side
effects outside of this code.  All this change is meant to do is wrap the
container_of() in the list iterator. When using the _entry_ variants
it is wrapping container_of() anyway so just going one level further
up the hierarchy of nested structures.

struct a {
	struct b {
		struct list_head l;
	}
}

It's actually a list of struct a as all elements on this list are
struct b instances within struct a, but you are treating it as a list
of struct b and then using a container_of() to get to struct a on
each one.

The change is treat it as a list of struct a with the list_head happening
to be wrapped in struct b.  Results in slightly simpler code and makes
the point these are always struct a instances.

Jonathan



> 
> 
> >> +
> >> +		if (cpumask_test_cpu(cpu, &dom->ctrl_comp->affinity))
> >> +			return dom;
> >> +	}
> >> +
> >> +	return NULL;
> >> +}
> >> +